Storage and calculation integrated device

Through the design of three-dimensional stacking and memory integrated modules of logic chips and DRAM units, the problem that existing memory integrated devices are difficult to achieve high capacity and high performance, and a memory integrated device with high integration and low power consumption is realized. Meet the needs of artificial intelligence computing power.

CN119179673AActive Publication Date: 2024-12-24SHANGHAI XINGAOFENG MICROELECTRONICS CO LTD
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202411189030.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-27
Publication Date
2024-12-24
Estimated Expiration
2044-08-27

AI Technical Summary

Technical Problem

The existing storage and computing integrated devices are difficult to achieve high capacity, high performance, and high integration storage and computing integration, and cannot meet the needs of high computing power and energy efficiency ratio of artificial intelligence computing power.

Method used

Design a memory and computing integrated device, through a three-dimensional stacking of logic chips and multiple DRAM units, combined with memory and computing integrated modules and DRAM control logic, realize high-capacity and short-distance connection, and dynamically select memory and computing integrated modules and arbitration logic. DRAM control logic, optimizes signal paths, and supports the interconnection of three-dimensional and two-dimensional/two-point five-dimensional interfaces.

Benefits of technology

Without increasing space occupation, the high performance and large capacity of the memory and computing integrated device are realized, simplifying the sharing of DRAM units by the system on-chip and memory integrated modules, supporting capacity up to tens of terabytes, replacing traditional DDR SDRAM , reduce power consumption and improve performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119179673A_ABST
    Figure CN119179673A_ABST
Patent Text Reader

Abstract

The embodiment of the invention provides a storage and calculation integrated device. The storage and calculation integrated device comprises at least one system-on-chip, a logic chip and a plurality of stacked DRAM (Dynamic Random Access Memory) units, the logic chip comprises a plurality of first-class interfaces which are interconnected with at least one system-on-chip; the plurality of storage and calculation integrated modules correspond to and are interconnected with one first-class interface; the plurality of DRAM control logics correspond to the plurality of storage and calculation integrated modules; the plurality of second-class interfaces correspond to the plurality of DRAM control logics; each DRAM control logic is interconnected with at least one DRAM unit through a plurality of second type interfaces; the arbitration logic corresponds to and is interconnected with one first-class interface, and is interconnected with the plurality of storage and calculation integrated modules and the plurality of DRAM control logics; the at least one system-on-chip is configured to access at least one of the plurality of corresponding storage and calculation integrated modules through the at least one first-class interface; and / or accessing at least one of the plurality of corresponding DRAM control logics through the at least one first type interface and the corresponding arbitration logic.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of semiconductor technology, and relates to but is not limited to a storage and computing integrated device. Background Art

[0002] Processing In Memory (PIM) technology is a hot topic in the industry and a development trend of artificial intelligence (AI). The demand for AI computing power has led to the ultra-high computing power and energy efficiency of PIM devices, which requires large capacity, high performance, and high integration of PIM devices. Summary of the invention

[0003] In view of this, an embodiment of the present application provides a storage and computing integrated device.

[0004] In a first aspect, an embodiment of the present application provides a storage-computing integrated device, which includes at least one system on chip, a logic chip interconnected with the at least one system on chip, and a plurality of DRAM units stacked on the logic chip and interconnected with the logic chip; wherein the logic chip includes: a plurality of first-class interfaces interconnected with at least one system on chip; a plurality of storage-computing integrated modules corresponding to and interconnected with one first-class interface; a plurality of DRAM control logics corresponding to the plurality of storage-computing integrated modules; a plurality of second-class interfaces corresponding to the plurality of DRAM control logics; each DRAM control logic is interconnected with at least one DRAM unit through a plurality of second-class interfaces; an arbitration logic corresponding to and interconnected with one first-class interface, and is interconnected with a plurality of storage-computing integrated modules and a plurality of DRAM control logics; at least one system on chip is configured to access at least one of the corresponding plurality of storage-computing integrated modules through at least one first-class interface; and / or to access at least one of the corresponding plurality of DRAM control logics through at least one first-class interface and the corresponding arbitration logic.

[0005] In some embodiments, the logic chip also includes an inter-interface interconnection connecting multiple first-class interfaces; at least one system on chip is configured to connect at least another first-class interface through a first-class interface and the inter-interface interconnection, and access at least one of the multiple storage-computing modules and / or at least one of the multiple DRAM control logics corresponding to at least another first-class interface.

[0006] In some embodiments, at least one system on chip is configured to access at least one of the corresponding multiple storage-computing integrated modules through a first-class interface, or to access at least one of the multiple storage-computing integrated modules corresponding to another first-class interface through a first-class interface and another first-class interface connected by an inter-interface interconnection; at least one of the multiple storage-computing integrated modules accessed is configured as a cache device; or is configured as a computing device to access at least one of the multiple DRAM control logics through arbitration logic.

[0007] In some embodiments, at least one system on chip is configured to: access a first-class interface; decode a corresponding first-class interface based on an address transmitted by at least one system on chip; select at least one of the corresponding multiple storage-computing integrated modules through a corresponding first-class interface for command and data transmission, or select at least one of the multiple storage-computing integrated modules corresponding to another first-class interface through a corresponding first-class interface and an interconnection between interfaces for command and data transmission; at least one of the multiple storage-computing integrated modules accessed starts calculation after receiving commands and data; during calculation, access at least one of the multiple DRAM control logics through arbitration logic, or decode to a corresponding first-class interface, connect another first-class interface through a corresponding first-class interface and an interconnection between interfaces, and access at least one of the multiple DRAM control logics corresponding to another first-class interface; after the calculation is completed, return the result and status to at least one system on chip through the original path.

[0008] In some embodiments, at least one system on chip is configured to access at least one of multiple DRAM control logics through a first-class interface and corresponding arbitration logic, or to access at least one of the multiple DRAM control logics through a first-class interface and another first-class interface corresponding to the inter-interface interconnection, and through another first-class interface and another corresponding arbitration logic; at least one of the multiple DRAM control logics accessed is configured to access the corresponding at least one DRAM unit.

[0009] In some embodiments, at least one system on chip is configured to: access a first-class interface; a corresponding first-class interface decodes based on an address transmitted by at least one system on chip; through a corresponding first-class interface, select at least one of the corresponding multiple DRAM control logics to transmit commands and data, or, through a corresponding first-class interface and another first-class interface connected by an inter-interface interconnection, select at least one of the multiple DRAM control logics corresponding to another first-class interface to transmit commands and data; after at least one of the multiple DRAM control logics accessed receives the command and data, it sends it to the corresponding multiple second-class interfaces after scheduling; the corresponding multiple second-class interfaces transmit the command and data to at least one DRAM unit; at least one DRAM unit completes the corresponding function; after at least one of the multiple DRAM control logics accessed completes the command, it returns the result and status to at least one system on chip through the original path.

[0010] In some embodiments, at least one system on chip is configured to access at least one of the corresponding multiple storage-computing integrated modules through a first-class interface, or to access at least one of the multiple storage-computing integrated modules corresponding to another first-class interface through a first-class interface and another first-class interface connected to the interface interconnection; at least one of the multiple storage-computing integrated modules accessed is configured as a computing device, and accesses at least one of the multiple DRAM control logics through arbitration logic; at least one system on chip is configured to access at least one of the multiple DRAM control logics through a first-class interface and the corresponding arbitration logic, or to access at least one of the multiple DRAM control logics through a first-class interface and another first-class interface connected to the interface interconnection, and through another first-class interface and another corresponding arbitration logic; wherein the arbitration logic is configured to select at least one of the multiple storage-computing integrated modules or at least one system on chip to access at least one of the multiple DRAM control logics through arbitration dynamic selection / software configured static selection.

[0011] In some embodiments, at least one system on chip is configured to: access a first-class interface; a corresponding first-class interface is decoded based on an address transmitted by at least one system on chip; through a corresponding first-class interface, at least one of the corresponding multiple storage-computing integrated modules and at least one of the corresponding multiple DRAM control logics are selected for command and data transmission, or, through a corresponding first-class interface and another first-class interface connected by an interconnection between interfaces, at least one of the multiple storage-computing integrated modules corresponding to another first-class interface and at least one of the corresponding multiple DRAM control logics are selected for command and data transmission through arbitration dynamic selection / software configuration static selection, and at least one of the multiple storage-computing integrated modules or at least one system on chip accesses at least one of the multiple DRAM control logics; after receiving the command and data, at least one of the accessed multiple DRAM control logics sends it to the corresponding multiple second-class interfaces after scheduling; the corresponding multiple second-class interfaces transmit the command and data to at least one DRAM unit; at least one DRAM unit completes the corresponding function; after at least one of the accessed multiple DRAM control logics completes the command, it returns the result and status to at least one system on chip through the original path.

[0012] In some embodiments, the logic chip further includes a three-dimensional interface, a two-dimensional / two-and-a-half-dimensional interface, and a path selection circuit; at least one system on chip is configured to: connect the three-dimensional interface to a first path of a first type of interface through the path selection circuit; the first path is used for three-dimensional interconnection between the logic chip and the system on chip; or, connect the two-dimensional / two-and-a-half-dimensional interface to a second path of the first type of interface through the path selection circuit; the second path is used for two-dimensional / two-and-a-half-dimensional interconnection between the logic chip and the system on chip.

[0013] In some embodiments, the logic chip is configured to: initialize and complete chip configuration; the chip configuration includes defining a first path or a second path, selecting the system on chip to connect to the corresponding first type interface, and allowing the system on chip to access the corresponding first type interface through the first path or the second path.

[0014] In some embodiments, the logic chip also includes a miscellaneous device\miscellaneous device dimension interface.

[0015] In a second aspect, an embodiment of the present application provides a storage-computing integrated device, which includes at least one stacked system on chip, a logic chip interconnected with the at least one system on chip, and a plurality of DRAM units stacked on the logic chip and interconnected with the logic chip; wherein the logic chip includes: a plurality of first-class interfaces interconnected with at least one system on chip; a plurality of storage-computing integrated modules corresponding to and interconnected with one first-class interface; a plurality of DRAM control logics corresponding to the plurality of storage-computing integrated modules; a plurality of second-class interfaces corresponding to the plurality of DRAM control logics; each DRAM control logic is interconnected with at least one DRAM unit through a plurality of second-class interfaces; an arbitration logic corresponding to and interconnected with one first-class interface, and is interconnected with a plurality of storage-computing integrated modules and a plurality of DRAM control logics; wherein a plurality of storage-computing integrated modules, a plurality of DRAM control logics and an arbitration logic corresponding to a first-class interface are arranged flat; at least one system on chip is configured to access at least one of the corresponding plurality of storage-computing integrated modules through at least one first-class interface; and / or to access at least one of the corresponding plurality of DRAM control logics through at least one first-class interface and a corresponding arbitration logic.

[0016] In some embodiments, the logic chip also includes an inter-interface interconnection connecting multiple first-class interfaces; at least one system on chip is configured to connect at least another first-class interface through a first-class interface and the inter-interface interconnection, and access at least one of the multiple storage-computing modules and / or at least one of the multiple DRAM control logics corresponding to at least another first-class interface.

[0017] In some embodiments, at least one system on chip is configured to access at least one of the corresponding multiple storage-computing integrated modules through a first-class interface, or to access at least one of the multiple storage-computing integrated modules corresponding to another first-class interface through a first-class interface and another first-class interface connected by an inter-interface interconnection; at least one of the multiple storage-computing integrated modules accessed is configured as a cache device; or is configured as a computing device to access at least one of the multiple DRAM control logics through arbitration logic.

[0018] In some embodiments, at least one system on chip is configured to access at least one of multiple DRAM control logics through a first-class interface and corresponding arbitration logic, or to access at least one of the multiple DRAM control logics through a first-class interface and another first-class interface corresponding to the inter-interface interconnection, and through another first-class interface and another corresponding arbitration logic; at least one of the multiple DRAM control logics accessed is configured to access the corresponding at least one DRAM unit.

[0019] In some embodiments, at least one system on chip is configured to access at least one of the corresponding multiple storage-computing integrated modules through a first-class interface, or to access at least one of the multiple storage-computing integrated modules corresponding to another first-class interface through a first-class interface and another first-class interface connected to the interface interconnection; at least one of the multiple storage-computing integrated modules accessed is configured as a computing device, and accesses at least one of the multiple DRAM control logics through arbitration logic; at least one system on chip is configured to access at least one of the multiple DRAM control logics through a first-class interface and the corresponding arbitration logic, or to access at least one of the multiple DRAM control logics through a first-class interface and another first-class interface connected to the interface interconnection, and through another first-class interface and another corresponding arbitration logic; wherein the arbitration logic is configured to select at least one of the multiple storage-computing integrated modules or at least one system on chip to access at least one of the multiple DRAM control logics through arbitration dynamic selection / software configured static selection.

[0020] In some embodiments, the logic chip further includes a three-dimensional interface, a two-dimensional / two-and-a-half-dimensional interface, and a path selection circuit; at least one system on chip is configured to: connect the three-dimensional interface to a first path of a first type of interface through the path selection circuit; the first path is used for three-dimensional interconnection between the logic chip and the system on chip; or, connect the two-dimensional / two-and-a-half-dimensional interface to a second path of the first type of interface through the path selection circuit; the second path is used for two-dimensional / two-and-a-half-dimensional interconnection between the logic chip and the system on chip.

[0021] In each embodiment of the present application, a three-dimensional stacking of a logic chip and a plurality of DRAM units is realized, and the space occupation is reduced while the performance of the storage-computing integrated device is less affected. The storage-computing integrated module is a computing unit (including a storage space), and the storage-computing integrated module realizes a high-capacity, short-distance connection through a DRAM controller and a DRAM unit to complete the storage-computing integrated calculation. At the same time, the DRAM unit can be used by the storage-computing integrated module, and the system on chip can also use DRAM alone through the first type of interface (SOC interface) or interact with the storage-computing integrated module through the DRAM unit. The system on chip can also access the storage space in the storage-computing integrated module through the SOC interface and interact directly with the storage-computing integrated module. In other words, the logic chip includes a storage-computing integrated module, and is interconnected with the DRAM unit in three dimensions to realize the acceleration effect of the storage-computing integrated module; the system on chip accesses the DRAM unit through the logic chip, and the computing integrated module is used as a DRAM unit; the logic chip contains DRAM control logic, which simplifies the sharing of DRAM units by the system on chip and the storage-computing integrated module; and the large capacity, high performance and high integration of the storage-computing integrated device are realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1A This is one of the cross-sectional schematic diagrams of the stacked storage and computing integrated device according to various embodiments of the present application;

[0023] Figure 1B This is a second cross-sectional schematic diagram of a stacked storage and computing integrated device according to various embodiments of the present application;

[0024] Figure 2 This is a third cross-sectional schematic diagram of a stacked storage-computing integrated device according to various embodiments of the present application;

[0025] Figure 3 This is a fourth cross-sectional schematic diagram of a stacked storage and computing integrated device according to various embodiments of the present application;

[0026] Figure 4 One of the schematic block diagrams of various modules of the logic chip of the storage-computing integrated device according to various embodiments of the present application;

[0027] Figure 5 The second schematic block diagram of each module of the logic chip of the storage-computing integrated device according to each embodiment of the present application;

[0028] Figure 6 This is the third schematic block diagram of each module of the logic chip of the storage and computing integrated device according to each embodiment of the present application. DETAILED DESCRIPTION

[0029] The following is a clear and complete description of the technical solutions in the embodiments of the present application in combination with the embodiments of the present application and the accompanying drawings. The described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0030] In the following description, a large number of specific details are given to provide a more thorough understanding of the present application. However, it is obvious to those skilled in the art that the present application can be implemented without one or more of these details. In other examples, in order to avoid confusion with the present application, some technical features known in the art are not described; that is, all features of the actual embodiments are not described here, and well-known functions and structures are not described in detail.

[0031] In the drawings, the sizes of layers, regions, elements and their relative sizes may be exaggerated for clarity. Like reference numerals denote like elements throughout.

[0032] It should be understood that when an element or layer is referred to as "on ...", "adjacent to ...", "connected to" or "coupled to" other elements or layers, it can be directly on, adjacent to, connected to or coupled to other elements or layers, or there can be intervening elements or layers. On the contrary, when an element is referred to as "directly on ...", "directly adjacent to ...", "directly connected to" or "directly coupled to" other elements or layers, there is no intervening element or layer. It should be understood that although the terms first, second, third, etc. can be used to describe various elements, components, regions, layers and / or parts, these elements, components, regions, layers and / or parts should not be limited by these terms. These terms are only used to distinguish one element, component, region, layer or part from another element, component, region, layer or part. Therefore, without departing from the teachings of the present application, the first element, component, region, layer or part discussed below can be represented as the second element, component, region, layer or part. And when the second element, component, region, layer or part is discussed, it does not indicate that the present application necessarily has the first element, component, region, layer or part.

[0033] Spatially relative terms such as "under", "beneath", "below", "under", "above", "above", etc., may be used here for convenience of description to describe the relationship between an element or feature shown in the figure and other elements or features. It should be understood that in addition to the orientation shown in the figure, the spatial relationship terms are intended to also include different orientations of the device in use and operation. For example, if the device in the accompanying drawings is turned over, then the elements or features described as "under other elements" or "under it" or "under it" will be oriented as "on" other elements or features. Therefore, the exemplary terms "under" and "under" may include both upper and lower orientations. The device can be oriented otherwise (rotated 90 degrees or other orientations) and the spatial descriptors used herein are interpreted accordingly.

[0034] The purpose of the terms used herein is only to describe specific embodiments and is not intended to be limiting of the present application. When used herein, the singular forms "one", "an" and "said / the" are also intended to include plural forms, unless the context clearly indicates otherwise. It should also be understood that the terms "consisting of" and / or "comprising", when used in this specification, determine the presence of the features, integers, steps, operations, elements and / or parts, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, parts and / or groups. When used herein, the term "and / or" includes any and all combinations of the relevant listed items.

[0035] In order to thoroughly understand the present application, detailed steps and detailed structures will be presented in the following description to illustrate the technical solution of the present application. The preferred embodiments of the present application are described in detail below, but in addition to these detailed descriptions, the present application may also have other implementation methods.

[0036] refer to Figure 1A and Figure 4 , Figure 2 and Figure 4 ,or Figure 3 and Figure 4 In a first aspect, an embodiment of the present application provides a storage-computing integrated device, which includes at least one system on chip 100, a logic chip 200 interconnected with the at least one system on chip, and a plurality of DRAM units 300 stacked on the logic chip and interconnected with the logic chip; wherein the logic chip 200 includes: a plurality of first-class interfaces interconnected with at least one system on chip; a plurality of storage-computing integrated modules corresponding to and interconnected with one first-class interface; a plurality of DRAM control logics corresponding to the plurality of storage-computing integrated modules; a plurality of second-class interfaces corresponding to the plurality of DRAM control logics; each DRAM control logic is interconnected with at least one DRAM unit through a plurality of second-class interfaces; an arbitration logic corresponding to and interconnected with one first-class interface, and is interconnected with a plurality of storage-computing integrated modules and a plurality of DRAM control logics; at least one system on chip is configured to access at least one of the corresponding plurality of storage-computing integrated modules through at least one first-class interface; and / or to access at least one of the corresponding plurality of DRAM control logics through at least one first-class interface and the corresponding arbitration logic.

[0037] For the convenience of describing the present application, the first direction and the second direction in each embodiment of the present application are represented as two orthogonal directions extending from the top / bottom surface of the DRAM (Dynamic Random Access Memory) unit, which can also be understood as two orthogonal directions parallel to the top / bottom surface of the DRAM unit; the third direction is a direction perpendicular to the top / bottom surface of the DRAM unit, which can be understood as the stacking direction of multiple DRAM units. The first direction is represented as the X direction in the accompanying drawings; the second direction is represented as the Y direction in the accompanying drawings; and the third direction is represented as the Z direction in the accompanying drawings.

[0038] To facilitate the description of the present application, the terms in the embodiments of the present application, such as "first-class interface", "second-class interface", "storage and computing integrated module", "DRAM control logic", and "arbitration logic" can also be understood by referring to the "SOC interface", "DRAM interface", "PIM", "DMC", and "Ari" shown in the accompanying drawings, respectively.

[0039] At least one system-on-chip (SOC) can access multiple DRAM units in a substantially simultaneous / non-simultaneous manner through at least one of the multiple first-type interfaces 201, 202, 203, 204. In some embodiments, at least one system-on-chip can access multiple DRAM units in a substantially simultaneous manner through at least a plurality of the multiple first-type interfaces 201, 202, 203, 204 in parallel. In some embodiments, at least one system-on-chip can include multiple stacked systems-on-chip.

[0040] A plurality of DRAM cells may include 4 stacked layers, and the capacity may support 12 gigabytes (GB) to 16GB, but the capacity is not limited to this range and may be less than 12GB or greater than 16GB. A plurality of DRAM cells may also include any number of stacked layers lower or higher than 4 layers.

[0041] The DRAM Memory Controller (DMC) can control the DRAM cells, such as write operations, read operations, and refresh operations on the DRAM cells. The DRAM control logic is configured to store data into the DRAM cells, read data stored in the DRAM cells, or refresh data stored in the DRAM cells.

[0042] The arbitration logic includes a multiplexer (MUX) and accesses the plurality of DRAM cells in a substantially simultaneous / non-simultaneous manner by dynamically / statically selecting at least one of the plurality of DRAM control logics.

[0043] The first type of interface may include a three-dimensional (3D) interface, for example, a through-silicon via (TSV) or a hybrid bonding structure having a vertical conductive channel; or, the first type of interface may include a two-dimensional (2D) / two-and-a-half-dimensional (2.5D) interface, for example, a physical layer (Physical Layer, PHY), a controller (CTRL). The number of the first type of interface may be 4, or any number lower or higher than 4, for example, refer to Figure 1A , Figure 2 or Figure 3 , schematically, the number of first-type interfaces is 4.

[0044] The second type of interface may include a 3D interface with a TSV or hybrid bonding structure having a vertical conductive channel. The 3D interface enables signals to pass through the layers of the logic chip 200 and multiple DRAM cells. In some embodiments, the 3D interface of the TSV or hybrid bonding structure may or may not pass through the logic chip to achieve interconnection between the DRAM and the logic chip.

[0045] refer to Figure 1A In some embodiments, the storage-computing integrated device includes at least one system on chip, a logic chip located on the at least one system on chip and interconnected with the at least one system on chip, and a plurality of DRAM units stacked on the logic chip and interconnected with the logic chip.

[0046] refer to Figure 1B In other embodiments, at least one logic chip interconnected with the system on chip, a plurality of DRAM units stacked on the logic chip and interconnected with the logic chip, and a system on chip located on the plurality of DRAM units and interconnected with the logic chip. In this way, the heat dissipation of the SOC in certain scenarios can be facilitated.

[0047] In the following description, reference is made to Figure 1A The storage and computing integrated device constructed in the stacking manner shown is described in detail, but is not used to limit the embodiments of the present application.

[0048] In some embodiments, a DRAM controller under multiple first-class interfaces connected to a system on chip can access all DRAM units in the corresponding physical space. For example, if multiple DRAM units are stacked in multiple layers, the DRAM controller can access one layer of DRAM units in the corresponding physical space, or multiple layers of DRAM units.

[0049] Schematically, multiple DRAM cells are stacked in 2 layers; each layer of DRAM cells includes 16 cell areas, corresponding to 16 DRAM control logics and 16 storage-computing integrated modules. Exemplarily, each cell area can be a DRAM accessible device (macro) or a storage body (BANK) (including multiple macros) of a minimum unit of a DRAM cell, and one storage body corresponds to one DRAM control logic and one storage-computing integrated module. Schematically, at least one system on chip 100 is interconnected with a logic chip 200 through four first-class interfaces 201, 202, 203, and 204, and the logic chip 200 is interconnected with two layers of DRAM cells through 16 second-class interfaces. Schematically, each of the four first-class interfaces 201, 202, 203, and 204 corresponds to an arbitration logic, four DRAM control logics, and four storage-computing integrated modules, and each of the four DRAM control logics corresponds to two cell areas, and the two cell areas are respectively located in two layers of DRAM cells.

[0050] It should be noted that the number of first-class interfaces and the number of second-class interfaces are used to generally describe the number of signal paths and are not associated with the specific number of pins. For example, at least one system on chip is interconnected with the logic chip through four first-class interfaces 201, 202, 203, and 204, which can indicate that at least one system on chip is interconnected with the logic chip through four paths (which can be four-way parallel interconnection), but does not mean that at least one system on chip is interconnected with the logic chip through four pins. For another example, a DRAM control logic is interconnected with a DRAM unit through two second-class interfaces, which can indicate that a DRAM control logic is interconnected with a DRAM unit through two paths (which can be two-way parallel interconnection), but does not mean that a DRAM control logic is interconnected with a DRAM unit through two pins.

[0051] In the embodiments of the present application, large capacity is achieved. For example, multiple DRAM units include 4 stacked layers, and the capacity can support 12GB to 16GB. The capacity is not limited to this range, and can also be less than 12GB or greater than 16GB, for example, up to tens of terabytes (TB), to meet capacity requirements, and replace, for example, Double Data Rate Synchronous Dynamic Random Access Memory (DDR SDRAM), which can remove the physical layer or reduce the power consumption of the physical layer, greatly improving the performance used by customers and reducing customer power consumption.

[0052] In some embodiments, at least one system on chip accesses a corresponding first-class interface 202; a corresponding first-class interface 202 is decoded based on an address transmitted by at least one system on chip; and through a corresponding first-class interface 202, at least one of the multiple storage-computing integrated modules and / or at least one of the multiple DRAM control logics corresponding to the first-class interface 202 is accessed to perform command and data transmission.

[0053] refer to Figure 5 In some embodiments, the logic chip 200 also includes an inter-interface interconnection connecting multiple first-class interfaces; at least one system on chip is configured to connect at least another first-class interface through a first-class interface and an inter-interface interconnection, and access at least one of the multiple storage-computing modules and / or at least one of the multiple DRAM control logics corresponding to at least another first-class interface.

[0054] In some embodiments, the inter-interface interconnect 206 may decode commands between system-on-chip interfaces and access a corresponding first-type interface 202 to another first-type interface 204 .

[0055] Exemplarily, at least one system on chip accesses a corresponding first-class interface 202; a corresponding first-class interface 202 is decoded based on an address transmitted by at least one system on chip; another first-class interface 204 is connected through a corresponding first-class interface 202 and an inter-interface interconnection 206, and another first-class interface 204 is decoded based on the address transmitted by the inter-interface interconnection, and accesses at least one of the multiple storage-computing integrated modules and / or at least one of the multiple DRAM control logics corresponding to the other first-class interface 204 to perform command and data transmission.

[0056] refer to Figure 5 In some embodiments, at least one system on chip is configured to access at least one of the corresponding multiple storage-computing integrated modules through a first-class interface 202, or to access at least one of the multiple storage-computing integrated modules corresponding to another first-class interface 204 by connecting another first-class interface 204 through a first-class interface 202 and an inter-interface interconnection 206; at least one of the multiple storage-computing integrated modules accessed is configured as a cache device; or is configured as a computing device to access at least one of the multiple DRAM control logics through arbitration logic.

[0057] In some embodiments, at least one system on chip may use one of the corresponding multiple storage-computing modules as a cache device or a computing device.

[0058] Exemplarily, at least one system on chip accesses a corresponding storage-computing integrated module 221 through a corresponding first-class interface 202; the accessed storage-computing integrated module 221 can be used as a cache device, and after the cache device completes the command, it returns the result and status to at least one system on chip through the original path; or, the accessed storage-computing integrated module 221 can be used as a computing device, and after the computing device completes the command, it returns the result and status to at least one system on chip through the original path; wherein, the process of the computing device completing the command will be described in detail later and will not be repeated here.

[0059] Exemplarily, at least one system on chip connects to another first-class interface 204 through a corresponding first-class interface 202 and an inter-interface interconnection 206, and accesses the storage-computing integrated module 241 corresponding to the other first-class interface 204. The process of the accessed storage-computing integrated module 241 completing the command can refer to the process of the accessed storage-computing integrated module 221 completing the command.

[0060] refer to Figure 5In some embodiments, at least one system on chip is configured to: access a first-class interface 202; a corresponding first-class interface 202 is decoded based on an address transmitted by at least one system on chip; through a corresponding first-class interface 202, at least one of the corresponding multiple storage-computing integrated modules (for example, storage-computing integrated module 221) is selected for command and data transmission, or, another first-class interface 204 is connected through a corresponding first-class interface 202 and an inter-interface interconnection 206, and at least one of the multiple storage-computing integrated modules corresponding to the other first-class interface 204 (for example, storage-computing integrated module 241) is selected for command and data transmission; after at least one of the multiple storage-computing integrated modules accessed receives the command and data, the operation starts; during the operation, at least one of the multiple DRAM control logics is accessed through the arbitration logic, or, after decoding to a corresponding first-class interface, another first-class interface is connected through a corresponding first-class interface and an inter-interface interconnection, and at least one of the multiple DRAM control logics corresponding to the other first-class interface is accessed; after the operation is completed, the result and status are returned to at least one system on chip through the original path.

[0061] Exemplarily, the accessed storage-computing module takes the storage-computing module 221 as an example. After receiving the command and data, the storage-computing module 221 starts the calculation; during the calculation, the storage-computing module 221 can access the DRAM control logic 222 through the arbitration logic 229; after the DRAM control logic 222 receives the command and data, it sends them to the corresponding DRAM interface after scheduling; the corresponding DRAM interface transmits the command and data to at least one DRAM unit; at least one DRAM unit completes the corresponding function; after the DRAM control logic 222 completes the command, it returns the result and status to the storage-computing module 221 through the original path; after the calculation is completed, the storage-computing module 221 returns the result and status to at least one on-chip system through the original path.

[0062] Exemplarily, the accessed storage-computing integrated module takes the storage-computing integrated module 221 as an example. After receiving the command and data, the storage-computing integrated module 221 starts the calculation; during the calculation, the storage-computing integrated module 221 can also decode to a corresponding first-class interface 202, connect to another first-class interface 204 through a corresponding first-class interface 202 and the interface interconnection 206, and access the DRAM control logic 242 through the arbitration logic 249 corresponding to the other first-class interface 204; after the DRAM control logic 242 receives the command and data, it sends it to the corresponding DRAM interface after scheduling; the corresponding DRAM interface transmits the command and data to at least one DRAM unit; at least one DRAM unit completes the corresponding function; after the DRAM control logic 242 completes the command, it returns the result and status to the storage-computing integrated module 221 through the original path; after the calculation is completed, the storage-computing integrated module 221 returns the result and status to at least one system on chip through the original path.

[0063] refer to Figure 5 In some embodiments, at least one system on chip is configured to access at least one of the multiple DRAM control logics through a first-class interface 202 and a corresponding arbitration logic 229, or to access at least one of the multiple DRAM control logics through a first-class interface 202 and an inter-interface interconnect 206 connected to a corresponding another first-class interface 204, and through another first-class interface 204 and a corresponding another arbitration logic 249; at least one of the multiple DRAM control logics accessed is configured to access at least one corresponding DRAM unit.

[0064] Exemplarily, at least one system on chip accesses the corresponding DRAM control logic 222 through a corresponding first-class interface 202 and a corresponding arbitration logic 229; the DRAM control logic 222 is accessed; after the DRAM control logic 222 receives the command and data, it sends it to the corresponding DRAM interface after scheduling; the corresponding DRAM interface transmits the command and data to at least one DRAM unit; at least one DRAM unit completes the corresponding function; after the DRAM control logic 222 completes the command, it returns the result and status to at least one system on chip through the original path.

[0065] Exemplarily, at least one system on chip is connected to another first-class interface 204 through a corresponding first-class interface 202 and an inter-interface interconnection 206, and accesses the corresponding DRAM control logic 242 through another first-class interface 204 and the corresponding arbitration logic 249. The process of the accessed storage-computing module 241 completing the command can refer to the process of the accessed storage-computing module 221 completing the command.

[0066] refer to Figure 5In some embodiments, at least one system on chip is configured to: access a first-class interface 202; a corresponding first-class interface 202 decodes based on an address transmitted by at least one system on chip; through a corresponding first-class interface 202, select at least one of the corresponding multiple DRAM control logics (for example, DRAM control logic 222) to transmit commands and data, or, connect another first-class interface 204 through a corresponding first-class interface 202 and an inter-interface interconnection 206, and select at least one of the multiple DRAM control logics corresponding to the other first-class interface 204 (for example, DRAM control logic 242) to transmit commands and data; at least one of the multiple DRAM control logics accessed (for example, DRAM control logic 222 / DRAM control logic 242) receives the command and data and sends it to the corresponding multiple second-class interfaces after scheduling; the corresponding multiple second-class interfaces transmit the command and data to at least one DRAM unit; at least one DRAM unit completes the corresponding function; after at least one of the multiple DRAM control logics accessed completes the command, it returns the result and status to at least one system on chip through the original path.

[0067] refer to Figure 5 In some embodiments, at least one system on chip is configured to access at least one of the corresponding multiple storage-computing modules through a first-class interface, or to access at least one of the multiple storage-computing modules corresponding to another first-class interface through a first-class interface and another first-class interface connected by an interconnection between interfaces; at least one of the multiple storage-computing modules accessed is configured as a computing device, and accesses at least one of the multiple DRAM control logics through arbitration logic; at least one system on chip is configured to access at least one of the multiple DRAM control logics through a first-class interface and a corresponding arbitration logic, or to access at least one of the multiple DRAM control logics through a first-class interface and another first-class interface connected by an interconnection between interfaces, and through another first-class interface and another corresponding arbitration logic; wherein the arbitration logic is configured to select at least one of the multiple storage-computing modules or at least one system on chip to access at least one of the multiple DRAM control logics through arbitration dynamic selection / software-configured static selection.

[0068] In some embodiments, the arbitration logic can judge or select the commands of the first type of interface and the commands of the storage-computing module. Exemplarily, the commands of the first type of interface and the commands of the storage-computing module can be accessed at the same time, and the arbitration logic can dynamically select through arbitration, or statically select through software configuration. For example, the arbitration logic can be used as a multiplexer (MUX), and the multiplexer can be configured by software to statically select the access of the commands of the first type of interface and the commands of the storage-computing module.

[0069] In some embodiments, the first type of interface can judge or select multiple commands of the storage-computing module. Exemplarily, multiple commands of the storage-computing module can be accessed by both at the same time, and the first type of interface can be dynamically selected through arbitration or statically selected through software configuration.

[0070] refer to Figure 5 In some embodiments, at least one system on chip is configured to: access a first-class interface 202; a corresponding first-class interface 202 is decoded based on an address transmitted by at least one system on chip; through a corresponding first-class interface 202, at least one of the corresponding multiple storage-computing integrated modules and at least one of the corresponding multiple DRAM control logics are selected for command and data transmission, or, through a corresponding first-class interface 202 and an inter-interface interconnection 206, another first-class interface 204 is connected, and at least one of the multiple storage-computing integrated modules corresponding to the other first-class interface 204 and at least one of the corresponding multiple DRAM control logics are selected for command and data transmission; the multiple storage-computing integrated modules accessed At least one of the modules and at least one of the multiple DRAM control logics being accessed, through dynamic arbitration selection / static software configuration selection when passing through the corresponding arbitration logic, one of the multiple storage-computing integrated modules being accessed or at least one on-chip system accesses at least one of the multiple DRAM control logics; after receiving the command and data, at least one of the multiple DRAM control logics being accessed sends it to the corresponding multiple second-class interfaces after scheduling; the corresponding multiple second-class interfaces transmit the command and data to at least one DRAM unit; at least one DRAM unit completes the corresponding function; after completing the command, at least one of the multiple DRAM control logics being accessed returns the result and status to at least one on-chip system through the original path.

[0071] Exemplarily, at least one system on chip selects the corresponding storage-computing integrated module 221 and the corresponding DRAM control logic 222 through a corresponding first-class interface 202 to transmit commands and data; wherein, the accessed storage-computing integrated module 221 accesses the DRAM control logic 222 through the arbitration logic 229 to transmit commands and data, and the first-class interface 202 corresponding to at least one system on chip also accesses the DRAM control logic 222 through the arbitration logic 229 to transmit commands and data; when passing through the corresponding arbitration logic 229, the arbitration logic 229 can judge or select the commands of the first-class interface and the commands of the storage-computing integrated module, and the arbitration dynamically selects / the software configures the static selection of the commands of the first-class interface or the commands of the storage-computing integrated module to access the DRAM control logic 222 for command and data transmission.

[0072] Exemplarily, at least one system on chip selects a corresponding storage-computing integrated module 221 and a corresponding DRAM control logic 222 through a corresponding first-class interface 202 to transmit commands and data; wherein, the storage-computing integrated module 221 is decoded to a corresponding first-class interface 202, and connected to another first-class interface 204 through a corresponding first-class interface 202 and an inter-interface interconnection 206 to transmit commands and data, and at least one system on chip also transmits commands and data through the corresponding first-class interface 202; when passing through the corresponding first-class interface 202, the first-class interface 202 can judge or select the commands of the storage-computing integrated module and the commands of at least one system on chip, and arbitrate dynamically select / software configure statically select the commands of the storage-computing integrated module or the commands of at least one system on chip for command and data transmission.

[0073] refer to Figure 6 In some embodiments, the logic chip further includes a three-dimensional interface, a two-dimensional / two-and-a-half-dimensional interface, and a path selection circuit; at least one system on chip is configured to: connect the three-dimensional interface to a first path of a first type of interface through the path selection circuit; the first path is used for three-dimensional interconnection between the logic chip and the system on chip; or, connect the two-dimensional / two-and-a-half-dimensional interface to a second path of the first type of interface through the path selection circuit; the second path is used for two-dimensional / two-and-a-half-dimensional interconnection between the logic chip and the system on chip.

[0074] refer to Figure 1A and Figure 6 In some embodiments, the 3D interface is used for 3D interconnection between the logic chip 200 and the system on chip 100, which can support a larger bandwidth. The 3D interface is used as a buffer interface of the system on chip 100. The 3D interface can be fixed in position, quantity, and size, and can be used as a standard interface or a coprocessor interface to quickly integrate different systems on chip.

[0075] refer to Figure 2 and Figure 6 In some embodiments, the 2D / 2.5D interface is used for 2D / 2.5D interconnection between the logic chip 200 and the system on chip 100 .

[0076] refer to Figure 3 and Figure 6In some embodiments, the 3D interface is used for 3D interconnection between the logic chip 200 and the system on chip 100, and the 2D / 2.5D interface is used for 2D / 2.5D interconnection between the logic chip 200 and the external system on chip 400, supporting compatibility between the 3D interface and the 2D / 2.5D interface. The 3D interface is used as a buffer interface of the system on chip 100; the 2D / 2.5D interface is used for docking with a 2D / 2.5D external system on chip 400. For example, if the external system on chip 400 includes an external buffer or a coprocessor, the 2D / 2.5D interface can be used as an interface for an external buffer of a 2D / 2.5D system on chip or a coprocessor interface of a storage-computing integrated module.

[0077] refer to Figure 6 In some embodiments, the path selection circuit includes a multiplexer; a first input end of the multiplexer is connected to the 3D interface, a second input end is connected to the 2D / 2.5D interface, an output end is connected to the SOC interface, and a control end receives a selection signal Sel to form a first path connecting the 3D interface to the first type interface or a second path connecting the 2D / 2.5D interface to the first type interface.

[0078] The SOC interface can be understood as the on-chip system convergence port of the 2D / 2.5D interface and the 3D interface. The actual path under different product forms is selected by selecting the signal sel, such as connecting the 3D interface to the first path of the first type of interface; or connecting the 2D / 2.5D interface to the second path of the first type of interface. The SOC interface can access the DRAM control logic or the storage and computing module. The SOC interface can also access the register space.

[0079] In some embodiments, the 3D interface may include a TSV or hybrid bonding structure having vertical conductive channels. The 3D interface enables signals to pass through various layers of the logic chip 200 and the plurality of DRAM cells.

[0080] In some embodiments, the 2D / 2.5D interface may include a physical layer, a controller (CTRL).

[0081] refer to Figure 6 In some embodiments, the logic chip 200 is configured to: initialize and complete chip configuration; the chip configuration includes defining a first path or a second path, selecting the system on chip to connect to the corresponding first type interface, and using the system on chip to access the corresponding first type interface through the first path or the second path.

[0082] Exemplarily, the logic chip 200 is initialized, chip configuration is completed, a 2D / 2.5D or 3D path is defined, and a SOC interface path is selected (which can be understood as selecting the first path or the second path).

[0083] After the system on chip 100 is selected through the 2D / 2.5D interface or the 3D interface, the SOC interface is accessed through the corresponding first path or the second path, so that the SOC interface performs decoding based on the address transmitted by the system on chip 100 .

[0084] refer to Figure 6 In some embodiments, the logic chip further includes a miscellaneous device\miscellaneous device dimension interface Misc.

[0085] In each embodiment of the present application, a 3D stacking of a logic chip and a plurality of DRAM units is achieved, and a specific memory capacity and a capacity-sensitive computing acceleration unit are integrated in the logic chip, and the unit and the DRAM unit are 3D stacked to form a storage-computing integrated module acceleration. When the storage-computing integrated module acceleration is not used, the capacity of the storage-computing integrated module can be used as DRAM by the system on chip, or when the storage-computing integrated module acceleration is used, part of the capacity of the storage-computing integrated module can be used by the system on chip. The multiple DRAM units included in the logic chip can be used as a cache of the system on chip, replacing, for example, DDR SDRAM, and can remove the physical layer or reduce the power consumption of the physical layer, thereby realizing large capacity, high performance, and high integration of the storage-computing integrated device.

[0086] In a second aspect, an embodiment of the present application provides another storage-computing integrated device, which includes at least one stacked system on chip, a logic chip interconnected with the at least one system on chip, and a plurality of DRAM units stacked on the logic chip and interconnected with the logic chip; wherein the logic chip includes: a plurality of first-class interfaces interconnected with at least one system on chip; a plurality of storage-computing integrated modules corresponding to and interconnected with one first-class interface; a plurality of DRAM control logics corresponding to a plurality of storage-computing integrated modules; a plurality of second-class interfaces corresponding to a plurality of DRAM control logics; each DRAM control logic is interconnected with at least one DRAM unit through a plurality of second-class interfaces; an arbitration logic corresponding to and interconnected with one first-class interface, and is interconnected with a plurality of storage-computing integrated modules and a plurality of DRAM control logics; wherein a plurality of storage-computing integrated modules, a plurality of DRAM control logics and an arbitration logic corresponding to a first-class interface are arranged flat; at least one system on chip is configured to access at least one of the corresponding plurality of storage-computing integrated modules through at least one first-class interface; and / or to access at least one of the corresponding plurality of DRAM control logics through at least one first-class interface and a corresponding arbitration logic.

[0087] In some embodiments, the logic chip also includes an inter-interface interconnection connecting multiple first-class interfaces; at least one system on chip is configured to connect at least another first-class interface through a first-class interface and the inter-interface interconnection, and access at least one of the multiple storage-computing modules and / or at least one of the multiple DRAM control logics corresponding to at least another first-class interface.

[0088] In some embodiments, at least one system on chip is configured to access at least one of the corresponding multiple storage-computing integrated modules through a first-class interface, or to access at least one of the multiple storage-computing integrated modules corresponding to another first-class interface through a first-class interface and another first-class interface connected by an inter-interface interconnection; at least one of the multiple storage-computing integrated modules accessed is configured as a cache device; or is configured as a computing device to access at least one of the multiple DRAM control logics through arbitration logic.

[0089] In some embodiments, at least one system on chip is configured to access at least one of multiple DRAM control logics through a first-class interface and corresponding arbitration logic, or to access at least one of the multiple DRAM control logics through a first-class interface and another first-class interface corresponding to the inter-interface interconnection, and through another first-class interface and another corresponding arbitration logic; at least one of the multiple DRAM control logics accessed is configured to access the corresponding at least one DRAM unit.

[0090] In some embodiments, at least one system on chip is configured to access at least one of the corresponding multiple storage-computing integrated modules through a first-class interface, or to access at least one of the multiple storage-computing integrated modules corresponding to another first-class interface through a first-class interface and another first-class interface connected to the interface interconnection; at least one of the multiple storage-computing integrated modules accessed is configured as a computing device, and accesses at least one of the multiple DRAM control logics through arbitration logic; at least one system on chip is configured to access at least one of the multiple DRAM control logics through a first-class interface and the corresponding arbitration logic, or to access at least one of the multiple DRAM control logics through a first-class interface and another first-class interface connected to the interface interconnection, and through another first-class interface and another corresponding arbitration logic; wherein the arbitration logic is configured to select at least one of the multiple storage-computing integrated modules or at least one system on chip to access at least one of the multiple DRAM control logics through arbitration dynamic selection / software configured static selection.

[0091] In some embodiments, the logic chip further includes a three-dimensional interface, a two-dimensional / two-and-a-half-dimensional interface, and a path selection circuit; at least one system on chip is configured to: connect the three-dimensional interface to a first path of a first type of interface through the path selection circuit; the first path is used for three-dimensional interconnection between the logic chip and the system on chip; or, connect the two-dimensional / two-and-a-half-dimensional interface to a second path of the first type of interface through the path selection circuit; the second path is used for two-dimensional / two-and-a-half-dimensional interconnection between the logic chip and the system on chip.

[0092] The other identical / similar parts of the storage-computing integrated device provided in each embodiment of the second aspect of the present application refer to the storage-computing integrated device provided in each embodiment of the first aspect mentioned above. For the technical features not fully disclosed in the embodiments of the second aspect of the present application, please refer to the above embodiments for understanding, and no further details will be given here.

[0093] In each embodiment of the present application, 3D integration of a system on chip, a logic chip, and multiple DRAM units is implemented, and space occupancy is reduced while having little impact on the performance of the storage-computing integrated device. Multiple storage-computing integrated modules included in the logic chip can be used as caches of the system on chip, replacing, for example, DDR SDRAM, and can remove the physical layer or reduce the power consumption of the physical layer, thereby implementing large capacity, high performance, and high integration of the storage-computing integrated device; 3D integration of the system on chip, a logic chip, and multiple DRAM units is implemented, and a 3D interface is defined to make standard parts; the storage-computing integrated module can support a capacity of terabytes (TB) level and can be used for AI large model calculations; the logic chip supports through silicon via or macro bump output, and is integrated with the system on chip for AXI protocol, with a capacity support of, for example, 400GB, and face-to-face (F2F) packaging; it supports enabling (disabling) at least one of multiple storage-computing integrated modules; and it supports compatibility between 3D interfaces and 2D / 2.5D interfaces.

[0094] It should be understood that "one embodiment" or "an embodiment" mentioned throughout the specification means that specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in various embodiments of the present application, the size of the sequence number of the above-mentioned processes does not mean the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application. The above-mentioned sequence numbers of the embodiments of the present application are only for description and do not represent the advantages and disadvantages of the embodiments.

[0095] The above description is only a preferred implementation mode of the present application, and does not limit the protection scope of the present application. All equivalent structural changes made by using the contents of the present application specification and drawings under the application concept of the present application, or direct / indirect application in other related technical fields are included in the protection scope of the present application.

Claims

1. A storage and computing integrated device, characterized in that: The invention comprises at least one system on chip, a logic chip interconnected with the at least one system on chip, and a plurality of DRAM units stacked on the logic chip and interconnected with the logic chip; wherein the logic chip comprises: a plurality of first type interfaces interconnected with the at least one system on chip; A plurality of storage-computing integrated modules corresponding to and interconnected with one of the first-type interfaces; A plurality of DRAM control logics corresponding to the plurality of storage-computing integrated modules; A plurality of second-type interfaces corresponding to the plurality of DRAM control logics; each of the DRAM control logics is interconnected with at least one of the DRAM units via the plurality of second-type interfaces; An arbitration logic corresponding to and interconnected with one of the first-type interfaces, and interconnected with the multiple storage-computing modules and the multiple DRAM control logics; The at least one system on chip is configured to access at least one of the corresponding multiple storage-computing modules through at least one of the first-type interfaces; and / or to access at least one of the corresponding multiple DRAM control logics through at least one of the first-type interfaces and the corresponding arbitration logic.

2. The storage and computing integrated device according to claim 1, characterized in that: The logic chip also includes an inter-interface interconnection connecting the plurality of first-type interfaces; The at least one system on chip is configured to connect to at least another first-class interface through one of the first-class interfaces and the interconnection between the interfaces, and access at least one of the multiple storage-computing modules and / or at least one of the multiple DRAM control logics corresponding to at least another first-class interface.

3. The storage and computing integrated device according to claim 2, characterized in that: The at least one system on chip is configured to access at least one of the corresponding multiple storage-computing integrated modules through one of the first-type interfaces, or to access at least one of the multiple storage-computing integrated modules corresponding to another of the first-type interfaces through one of the first-type interfaces and another of the first-type interfaces connected through the inter-interface interconnection; At least one of the plurality of storage-computing-in-one modules being accessed is configured as a cache device; Alternatively, the device is configured as a computing device and accesses at least one of the plurality of DRAM control logics through the arbitration logic.

4. The storage and computing integrated device according to claim 3, characterized in that: The at least one system on chip is configured to: accessing an interface of the first type; A corresponding one of the first-type interfaces decodes based on an address transmitted by the at least one system-on-chip; Through a corresponding one of the first-type interfaces, at least one of the corresponding multiple storage-computing integrated modules is selected to transmit commands and data, or, Connecting another first-class interface through a corresponding first-class interface and the interconnection between the interfaces, selecting at least one of the multiple storage-computing integrated modules corresponding to the other first-class interface to transmit commands and data; After receiving the command and data, at least one of the accessed multiple storage-computing integrated modules starts computing; When the operation is performed, at least one of the plurality of DRAM control logics is accessed through the arbitration logic, or, after being decoded to a corresponding one of the first-class interfaces, another first-class interface is connected through a corresponding one of the first-class interfaces and the inter-interface interconnection, and at least one of the plurality of DRAM control logics corresponding to another first-class interface is accessed; After the operation is completed, the result and status are returned to the at least one system on chip through the original path.

5. The storage and computing integrated device according to claim 2, characterized in that: The at least one system on chip is configured to access at least one of the plurality of DRAM control logics through one of the first-type interfaces and the corresponding arbitration logic, or to access at least one of the plurality of DRAM control logics through one of the first-type interfaces and another of the first-type interfaces corresponding to the inter-interface interconnection connection, and through another of the first-type interfaces and another of the corresponding arbitration logic; At least one of the accessed plurality of DRAM control logics is configured to access a corresponding at least one of the DRAM cells.

6. The storage and computing integrated device according to claim 5, characterized in that: The at least one system on chip is configured to: accessing an interface of the first type; A corresponding one of the first-type interfaces decodes based on an address transmitted by the at least one system-on-chip; Selecting at least one of the corresponding plurality of DRAM control logics to transmit commands and data through a corresponding one of the first-type interfaces, or, Connecting another first-class interface through a corresponding one of the first-class interfaces and the interconnection between the interfaces, and selecting at least one of the plurality of DRAM control logics corresponding to the other first-class interface to transmit commands and data; After receiving the command and data, at least one of the accessed multiple DRAM control logics sends the command and data to the corresponding multiple second-type interfaces after scheduling; The corresponding plurality of second-type interfaces transmit commands and data to at least one of the DRAM units; and at least one of the DRAM units performs a corresponding function; After completing the command, at least one of the accessed plurality of DRAM control logics returns the result and status to the at least one on-chip system through the original path.

7. The storage and computing integrated device according to claim 2, characterized in that: The at least one system on chip is configured to access at least one of the corresponding multiple storage-computing integrated modules through one of the first-type interfaces, or to access at least one of the multiple storage-computing integrated modules corresponding to another of the first-type interfaces through one of the first-type interfaces and another of the first-type interfaces connected through the inter-interface interconnection; At least one of the multiple storage-computing modules being accessed is configured as a computing device, and accesses at least one of the multiple DRAM control logics through the arbitration logic; The at least one system on chip is configured to access at least one of the plurality of DRAM control logics through one of the first-type interfaces and the corresponding arbitration logic, or to access at least one of the plurality of DRAM control logics through one of the first-type interfaces and another of the first-type interfaces corresponding to the inter-interface interconnection connection, and through another of the first-type interfaces and another of the corresponding arbitration logic; The arbitration logic is configured to select at least one of the multiple storage-computing modules or at least one on-chip system to access at least one of the multiple DRAM control logics through dynamic selection by arbitration / static selection by software configuration.

8. The storage and computing integrated device according to claim 7, characterized in that: The at least one system on chip is configured to: accessing an interface of the first type; A corresponding one of the first-type interfaces decodes based on an address transmitted by the at least one system-on-chip; Through a corresponding one of the first-type interfaces, select at least one of the corresponding multiple storage-computing integrated modules and at least one of the corresponding multiple DRAM control logics to perform command and data transmission, or, Connecting another first-class interface through a corresponding first-class interface and the interconnection between the interfaces, selecting at least one of the multiple storage-computing integrated modules and at least one of the corresponding multiple DRAM control logics corresponding to the other first-class interface to transmit commands and data; At least one of the multiple storage-computing integrated modules being accessed and at least one of the multiple DRAM control logics being accessed selects at least one of the multiple storage-computing integrated modules or at least one of the at least one system-on-chip to access at least one of the multiple DRAM control logics through arbitration dynamic selection / software configuration static selection when passing through the corresponding arbitration logic; After receiving the command and data, at least one of the accessed multiple DRAM control logics sends the command and data to the corresponding multiple second-type interfaces after scheduling; The corresponding plurality of second-type interfaces transmit commands and data to at least one of the DRAM units; and at least one of the DRAM units performs a corresponding function; After completing the command, at least one of the accessed plurality of DRAM control logics returns the result and status to the at least one on-chip system through the original path.

9. The storage and computing integrated device according to claim 1 or 2, characterized in that: The logic chip further includes a three-dimensional interface, a two-dimensional / two-and-a-half-dimensional interface, and a path selection circuit; the at least one system on chip is configured as: Connecting the three-dimensional interface to a first path of the first type of interface through the path selection circuit; the first path is used for three-dimensional interconnection between the logic chip and the system on chip; or, The two-dimensional / two-and-a-half-dimensional interface is connected to a second path of the first type of interface through the path selection circuit; the second path is used for two-dimensional / two-and-a-half-dimensional interconnection between the logic chip and the system on chip.

10. The storage and computing integrated device according to claim 9, characterized in that: The logic chip is configured as: Initialization completes chip configuration; the chip configuration includes defining the first path or the second path, selecting the system on chip to connect to the corresponding first type interface, and allowing the system on chip to access the corresponding first type interface through the first path or the second path.

11. The storage and computing integrated device according to claim 9, characterized in that: The logic chip also includes miscellaneous device\miscellaneous device dimension interface.

12. A storage and computing integrated device, characterized in that: The invention comprises at least one stacked system on chip, a logic chip interconnected with the at least one system on chip, and a plurality of DRAM units stacked on the logic chip and interconnected with the logic chip; wherein the logic chip comprises: a plurality of first type interfaces interconnected with the at least one system on chip; A plurality of storage-computing integrated modules corresponding to and interconnected with one of the first-type interfaces; A plurality of DRAM control logics corresponding to the plurality of storage-computing integrated modules; A plurality of second-type interfaces corresponding to the plurality of DRAM control logics; each of the DRAM control logics is interconnected with at least one of the DRAM units via the plurality of second-type interfaces; An arbitration logic corresponding to and interconnected with one of the first-type interfaces, and interconnected with the multiple storage-computing modules and the multiple DRAM control logics; Wherein, the multiple storage-computing integrated modules, the multiple DRAM control logics and the arbitration logic corresponding to one of the first-type interfaces are arranged in a flat pattern; The at least one system on chip is configured to access at least one of the corresponding multiple storage-computing modules through at least one of the first-type interfaces; and / or to access at least one of the corresponding multiple DRAM control logics through at least one of the first-type interfaces and the corresponding arbitration logic.

13. The storage and computing integrated device according to claim 12, characterized in that: The logic chip also includes an inter-interface interconnection connecting the plurality of first-type interfaces; The at least one system on chip is configured to connect to at least another first-class interface through one of the first-class interfaces and the interconnection between the interfaces, and access at least one of the multiple storage-computing modules and / or at least one of the multiple DRAM control logics corresponding to at least another first-class interface.

14. The storage and computing integrated device according to claim 13, characterized in that: The at least one system on chip is configured to access at least one of the corresponding multiple storage-computing integrated modules through one of the first-type interfaces, or to access at least one of the multiple storage-computing integrated modules corresponding to another of the first-type interfaces through one of the first-type interfaces and another of the first-type interfaces connected through the inter-interface interconnection; At least one of the plurality of storage-computing-in-one modules being accessed is configured as a cache device; Alternatively, the device is configured as a computing device and accesses at least one of the plurality of DRAM control logics through the arbitration logic.

15. The storage and computing integrated device according to claim 13, characterized in that: The at least one system on chip is configured to access at least one of the plurality of DRAM control logics through one of the first-type interfaces and the corresponding arbitration logic, or to access at least one of the plurality of DRAM control logics through one of the first-type interfaces and another of the first-type interfaces corresponding to the inter-interface interconnection connection, and through another of the first-type interfaces and another of the corresponding arbitration logic; At least one of the accessed plurality of DRAM control logics is configured to access a corresponding at least one of the DRAM cells.

16. The storage and computing integrated device according to claim 13, characterized in that: The at least one system on chip is configured to access at least one of the corresponding multiple storage-computing integrated modules through one of the first-type interfaces, or to access at least one of the multiple storage-computing integrated modules corresponding to another of the first-type interfaces through one of the first-type interfaces and another of the first-type interfaces connected through the inter-interface interconnection; At least one of the multiple storage-computing modules being accessed is configured as a computing device, and accesses at least one of the multiple DRAM control logics through the arbitration logic; The at least one system on chip is configured to access at least one of the plurality of DRAM control logics through one of the first-type interfaces and the corresponding arbitration logic, or to access at least one of the plurality of DRAM control logics through one of the first-type interfaces and another of the first-type interfaces corresponding to the inter-interface interconnection connection, and through another of the first-type interfaces and another of the corresponding arbitration logic; The arbitration logic is configured to select at least one of the multiple storage-computing modules or at least one on-chip system to access at least one of the multiple DRAM control logics through dynamic selection by arbitration / static selection by software configuration.

17. The storage and computing integrated device according to claim 12 or 13, characterized in that: The logic chip further includes a three-dimensional interface, a two-dimensional / two-and-a-half-dimensional interface, and a path selection circuit; the at least one system on chip is configured as: Connecting the three-dimensional interface to a first path of the first type of interface through the path selection circuit; the first path is used for three-dimensional interconnection between the logic chip and the system on chip; or, The two-dimensional / two-and-a-half-dimensional interface is connected to a second path of the first type of interface through the path selection circuit; the second path is used for two-dimensional / two-and-a-half-dimensional interconnection between the logic chip and the system on chip.

Citation Information

Patent Citations

  • Integrated chip integrating storage and calculation

    CN113626372A

  • Access controller

    CN116737617A

  • Device, chip, equipment, storage scheduling method and multi-layer neural network training method

    CN116976411A

  • Storage and calculation integrated chip architecture, packaging method and device

    CN117687955A

  • Storage and calculation integrated chip architecture, packaging method and device

    CN117708039A